I am trying to optimize a couple of molecular structures using DFT B3LYP/cc-pVDZ in solution (COSMO; epsilon=7.6).

I start by making a single-point calculation via dscf, followed by the geometry optimization via jobex. And finally want to to perform a frequency analysis using aoforce in order to verify that the found structure is indeed a minimum structure.

The problem is that it seems no matter what I do, I always end up with at least one imaginary frequency in the order of i10 cm^{-1}.

Here's what I have tried to get rid of that:

- Use stricter convergence criteria for the SCF energy and for the convergence criteria of the geometry optimization (stp submenu in define)

- Recalculate the hessian at every step in the optimization (did cause the calculation to not converge)

- Use a bigger grid for the DFT calculation (up to grid size of 7)

- Use screwer to distort the molecule along the imagainy mode

The molecule I have tried all these things on is an imino ester with 24 atoms (H, C, N, O).

Before starting to fiddle with all these parameters I (accidentally) performed a separate optimization using the BP86 functional that TurboMole suggests by default. With that I did not have any imaginary modes.

My question basically is: How can I get rid of these imaginary frequencies? Does this indicate that I should use a different functional than B3LYP?

EDIT: I am using Turbomole 7.4.1